Curated from knowledge graph (125 nodes, 145 edges) · All articles published within the last 7 days

• Anthropic adds persistent memory to Claude Managed Agents in public beta — Agents built on the Claude Platform can now learn and share knowledge across sessions; a direct unlock for TritonAI's agentic workflows and the Developer API Program, where session continuity has been a key gap. 🔗 Graph: anthropic, claude, agentic-ai, developer-api-program 📅 Published: 2026-04-23 📰 https://sdtimes.com/anthropic/anthropic-adds-memory-to-claude-managed-agents/

• Anthropic MCP Design Vulnerability Enables RCE, Threatening AI Supply Chain — OX Security disclosed 10 vulnerabilities (CVE-2026-30623) in Anthropic's MCP SDK stdio transport affecting LiteLLM, LangChain, LangFlow, and others; authenticated and unauthenticated command injection paths. Upgrade LiteLLM to v1.83.7-stable immediately — TritonAI's gateway is directly in scope. 🔗 Graph: litellm-enterprise, model-context-protocol, ai-security, tritongpt 📅 Published: 2026-04-21 📰 https://thehackernews.com/2026/04/anthropic-mcp-design-vulnerability.html

• OpenAI releases GPT-5.5, its "smartest and most intuitive" model yet — Positioned as a step toward an AI "super app," GPT-5.5 is more efficient and better at coding than GPT-5 and is rolling out to ChatGPT users and the API. Evaluate for TritonAI gateway routing as it may offer a cost/capability advantage over current GPT-5.4 defaults. 🔗 Graph: openai, gpt-4, litellm-enterprise, model-agnosticism 📅 Published: 2026-04-23 📰 https://techcrunch.com/2026/04/23/openai-chatgpt-gpt-5-5-ai-model-superapp/

• DeepSeek unveils V4 model tailored for Huawei chips as China pushes for tech autonomy — DeepSeek's 1.6-trillion-parameter V4 and a smaller V4-Flash variant trained on Huawei Ascend 950s demonstrate that frontier-quality models no longer require Nvidia hardware — with rock-bottom pricing that could reshape on-prem AI infrastructure planning at SDSC. 🔗 Graph: model-agnosticism, san-diego-supercomputer-center, tritongpt, ai-governance 📅 Published: 2026-04-24 📰 https://www.reuters.com/technology/chinas-deepseek-returns-with-new-model-year-after-viral-rise-2026-04-24/

• Google signs classified AI deal with Pentagon amid employee opposition — Google joined OpenAI and xAI in granting DoD rights to use its AI for "any lawful government purpose" in classified settings, even as hundreds of employees pushed back. Sets a tone for how major AI vendors will navigate government/national security contracts — relevant context for UC San Diego's own AI governance posture. 🔗 Graph: google, ai-governance, ai-compliance-governance, vertical-ai 📅 Published: 2026-04-28 📰 https://www.reuters.com/technology/google-signs-classified-ai-deal-with-pentagon-information-reports-2026-04-28/

• University of Wisconsin receives $100 million for its new AI college — UW–Madison named its founding dean and secured $100M in philanthropic support and 50 new faculty lines for its College of Computing and Artificial Intelligence, set to launch July 1. As UC San Diego positions TritonAI as a flagship, peer institutions formalizing AI at the college level raises the competitive bar. 🔗 Graph: higher-ed-ai, uc-san-diego, tritonai, ai-adoption 📅 Published: 2026-04-28 📰 https://universitybusiness.com/university-of-wisconsin-receives-100-million-for-its-new-ai-college/

• 5 AI Myths and Why We Must Move Past Them — Inside Higher Ed op-ed argues that standard advice on AI in teaching — including "tell students to disclose AI use" and "focus on detection" — is already outdated and is holding institutions back from genuine AI integration. Directly relevant to TritonAI's faculty adoption and the AI Symposium UCSD planning. 🔗 Graph: higher-ed-ai, tritonai, ai-adoption, ai-symposium-ucsd 📅 Published: 2026-04-28 📰 https://www.insidehighered.com/opinion/views/2026/04/28/5-ai-myths-and-why-we-must-move-past-them-opinion

💡 Signal: Two urgent items today. First: patch LiteLLM immediately — CVE-2026-30623 affects MCP stdio transport in your exact stack. Second: Anthropic's persistent memory for Managed Agents plus GPT-5.5's release represent back-to-back capability jumps that make the case for updating TritonAI's model routing strategy this week. On the governance front, Google's DoD deal and UW-Madison's $100M college launch signal that the institutional stakes around AI are escalating fast — useful framing for Brett's upcoming cabinet and conference presentations.